University of Michigan Linguistics & Philosophy
Workshop on Implicatures
November 21-23, 2008
Indirect answers and cooperation: On Asher and Lascarides’s
‘Making the right commitments in dialogue’
Christopher Potts
UMass Amherst
Abstract This commentary argues that linguistic cooperation is essential even
in discourse situations in which the nonlinguistic preferences of the participants
are misaligned. The central examples involve indirect answers to direct questions. The analysis builds on the work of Asher and Lascarides, without, though
retreating from the axioms of cooperativity as hastily as they do in the workshop
paper (Asher & Lascarides 2008). I also argue (section 4) that discourse coherence and inferences from the common ground can account for much pragmatic
enrichment.
1
The importance of coherence
For the semantically-inclined pragmaticist — the sort who wants to stick close
to language even while moving to the richer problems of message enrichment
in social contexts — reading Asher & Lascarides’s (2003) Logics of Conversation
(henceforth LoC) is inspiring. The overall message is that if you get the discourse
coherence relations right, and reason seriously in terms of contextual information, then enriched pragmatic meanings will emerge. The details are a grind, but
the basic idea is compelling enough to keep one’s spirits up.
I’ve only begun my study of LoC, but I see fairly clearly that cooperativity
is part of its foundation. When we walk back through the book’s important
high-level generalizations, we often encounter cooperativity along the way. The
discourse participants might want very different things in life, but, when they
exchange information in dialogue, they have an incentive to coordinate, not only
on the meanings of their words and phrases, but also on how those linguistic
pieces are woven into a discourse.
It’s thus surprising to find that ‘Making the right commitments in dialogue’
(henceforth ‘Commitments’) comes out boldly against this cooperative foundation. The abstract opens by saying that accounts like LoC’s “fail to say anything
about conversations where a robust notion of cooperation is absent”. The conversations A&L use to motivate this are extremely (even absurdly) confrontational
— so much so that it can be hard to imagine what they would involve except
unbridled aggression or unchecked duplicity.
1
Christopher Potts
Comments on Asher and Lascarides
With this commentary, I urge caution when moving away from the cooperative foundation. It is conceivable that A&L are right that most pragmatic theories
say little or nothing about situations so uncooperative that little information
can be exchanged. These might not be very interesting cases for pragmatic
theory, though. I argue that LoC and ‘Commitments’ have a lot to say about
situations in which preferences are misaligned but the players are still attempting
to communicate.
Indirect answers to direct questions are the empirical foundation of this
commentary. The next section introduces the central tool in LoC for modeling
such answers: the coherence relation of Indirect Question–Answer Pairs and its
associated axiom. I emphasize their connection to cooperativity as defined in
LoC. Section 3 reviews a range of indirect answer types, seeking, in particular, to
build a case that these relations, though rooted in cooperativity, are alive and
important even in discourse situations in which preferences are misaligned. As a
group, I think these cases support the LoC view of linguistic cooperation.
Section 4 then steps back to assess what else needs to be done in order to
deliver rich pragmatic inferences. I argue, with A&L, that the coherence relations,
working in conjunction with inferences from contextual information, tell much
of the story, and I review some very simple computational simulations to bolster
that conclusion. Finally, section 5 raises two tangentially related issues that
I personally would benefit from discussing with A&L and the other workshop
participants.
2
The Indirect Question–Answer Pair (IQAP) axiom
The Indirect Question–Answer Pair (IQAP) axiom, stated in (1), makes only a
brief appearance in ‘Commitments’, but it is central to LoC. I argue that it is
central to the concerns of ‘Commitments’ as well, though this is perhaps not
immediately evident.
(1)
IQAP: λ :?(α,β) ∧ int(α) > λ : IQAP(α,β)
‘Generally, if an interrogative α is related by some coherence relation ?
to β (in discourse λ), then that relation is the indirect question–answer
pair relation.’
IQAP establishes a defeasible connection between interrogatives and the
utterances that follow them in discourse: if you ask me a question α, and I
respond with β, then you will expect IQAP(α,β) to hold. I should, in turn, plan
my utterance with your expectations in mind. You are aware that I will plan with
these expectations in mind. I am aware that you’re aware of this. And so forth.
2
Christopher Potts
Comments on Asher and Lascarides
The result is stability in discourse coherence, albeit with room for exceptions.
In LoC, IQAP is not quite axiomatic. ‘Commitments’ doesn’t provide the
details of its derivation, but it hints at them when it says that IQAP and Q-Elab
“follow from their [LoC’s] formalization of axioms of rationality and cooperativity” (p. 11). It is worth expanding on this. In LoC, the primitive statement
is Cognitive IQAP (p. 404), which depends on the beliefs and inferential
capabilities of the questioner. A&L derive the simpler form (1) from Cognitive
IQAP and other axioms of their system. For present purposes, the step that
is most important is the one where they call upon Cooperativity (p. 391),
which is stated informally in (2).
(2)
Cooperativity
For two agents A and B in a single discourse:
a.
Generally, if A intends ϕ, then B intends ϕ.
b.
Generally, if A cannot adopt B’s intentions, then A intends to make
this known to B.
The derivation of IQAP from its cognitive counterpart is important because it
frees IQAP, in some sense, from its entanglements with deep pragmatic notions.
A&L write:
while validating these axioms is achieved through reasoning about
cognitive states (see Asher & Lascarides (2003)), the premises
of these axioms don’t talk about beliefs or intentions at all, and
instead rest entirely on sentence mood.
(p. 11)
I take this to mean the following: if we dig down deep inside of IQAP, we
find rich cognitive notions connected with cooperation, in the sense of shared
intentions and openness that (2) captures. IQAP inferences might seem to be
only form-based, but they are still mediated by genuinely pragmatic notions.
Thus, changing the pragmatics should change the IQAP inferences.
The next section assesses this idea. Before moving on, though, it is worth
briefly describing what the IQAP relation imposes on the connection between the
question α and the answer β. Here is a prose adaptation of the statement and
description in LoC (p. 403):1
(3)
IQAP(α,β) holds only if there is a true direct answer p to the question
¹αº, and the questioner can infer p from ¹βº in the utterance context.
1 I use ¹·º informally to pick out the interpretations of expressions.
3
Christopher Potts
Comments on Asher and Lascarides
I think some of the exchanges below call into question the emphasis on truth, but
the inferential relationship between the direct and indirect answers is important,
as is the orientation to the questioner’s capabilities.
3
Indirect answers and non-cognitive IQAP
My empirical focus is on question–answer pairs in which the answer is semantically (but not pragmatically) unrelated to the question. Such indirect answers are
central to the LoC account of the pragmatics of answerhood. A slight variation
on the following example opens their discussion of IQAP (p. 313):
(4)
a.
A: Did John pass his exams?
b.
B: He got 60%.
Let’s suppose that A’s question denotes the set {p,¬p}, where p is the proposition
that John passed his exams and ¬p is its complement. In general logical terms, B
does not fully address the issue raised by A: his answer, call it q, has a nonempty
intersection with both p and ¬p, which means that q fails to resolve the issue
(Groenendijk & Stokhof 1984; Roberts 1996; van Rooy 2004):2
(5)
A&L note that B seems not to have resolved A’s question. They continue, though,
by saying that A might well be able to reason in terms of what he knows in order
to turn B’s superficially partial answer into a resolving one. Suppose A knows
that a passing grade is at or above 60%. Then B has answered A’s question; in
(6), I’ve simply wiped out the space in which q ∧ ¬p is true, to reflect this world
knowledge.
2 It helps to imagine that we are simply unsure of the standard for failing. It could be 60%, but it
could be higher or lower.
4
Christopher Potts
Comments on Asher and Lascarides
(6)
Learning q brings us into the gray space. Because the upper right corner has been
eliminated, the gray space entails that John passed his exams. The necessary
condition in (3) is satisfied, because the questioner can calculate that β, together
with the common ground, entails the direct answer.
It is no surprise that answers like this can be resolving despite being superficially unrelated to the issue at hand. Speakers who use them are simply
implicitly asking their listeners to reason in terms of other knowledge they have.
It’s a common discourse strategy. In the Switchboard example (7), Speaker A.101
relies on it as well.3
(7)
B.100: So will, you, uh, breed one of the litter, then next year?
A.101: We’re, we sold all that litter,
One can’t breed what one has already sold. Thus, worlds in which A.101 has
both sold the entire litter and breeds one of its members are outside the realm
of possibility. As a result, Speaker B.100 likely felt that his question had been
thoroughly answered.
Of course, indirect answers of this form are not simply direct answers in
disguise. They are over-answers, i.e., they fully address the immediate question
and they also raise and answer some additional issue. For example, a simple
‘yes’ answer to A’s question in (4a) would take us to the left side of (6), but it
would leave open which numerical grade John got. In effect, B’s indirect answer
resolves both A’s question as well as the question of which numerical grade John
got (as well as all its subquestions, e.g., Did John get 60%?; see Büring 1999).
If this were all that indirect answers involved, then there would be little
need to invoke discourse coherence relations (or any of the Gricean stuff falling
under the heading of Relevance). These are logical deductions; the questioners
in (4) and (7) might as well have looked their answers up. In virtue of the
3 In a sample of 118 polar questions in the Switchboard Dialog Act Corpus (Jurafsky, Shriberg &
Biasca 1997), 23 are superficially partial answers, and at least 5 of those employ the indirect
strategy represented by (7). This seems like a healthy number in light of the fact that Switchboard
participants were strangers talking on the phone, and thus had relatively little contextual
knowledge to lean on.
5
Christopher Potts
Comments on Asher and Lascarides
way that these dialogues actually went, we will probably attribute some public
commitments to the answerers, and this might have consequences for how we
view their epistemic states, but the important work is done by the context.
However, not all indirect answers are as straightforward as these. Perhaps
the best way to see that is to swing all the way from the natural cases above
to more surprising dialogues like (8), which is a variation on an example that
Matthew Stone discusses in his commentary.
(8)
A: Did John take a train from Paris to Istanbul?
B: He likes spinach.
Let’s join A in her desperate struggle to figure out what B means. We have
learned that John likes spinach. By (3) and general principles of charitable
interpretation, we will assume that B assumes that there is an inferential path
from John’s taste for spinach to an answer to A’s question. Do we have access to
some world knowledge that, when fed this knowledge, will deliver a resolving
answer? Hard to say, hard to say. One thing is clear though: we are all of us
(A especially) working overtime to try to find a way to relate B’s reply to the
immediate question concerning John’s travel. As Stone says, “Our search for
coherence demands an interpretive connection, rationality or no”.
This is a source of speaker commitments to coherence relations, and thus it is
fertile ground for the theoretical approach that A&L take in their work. We might
never fully determine what B means to convey to us, and it might even be the
case that B wishes to throw us off the scent, as it were, by forcing us to consider
unimportant connections between spinach and rail travel. However, there is
little doubt that, in this situation, B will end up committed to the following
relationship (in the discourse situation of (8)):
(9)
IQAP(Did John take a train from Paris to Istanbul?,He likes spinach)
Two things about this are important:
(10)
a.
People are attuned to content like (9). It is the sort of thing that
we can debate, dispute, and query.
b.
Commitments of this form are robust across many kinds of discourse situation.
In their paper for the workshop, Franke, de Jager, and van Rooij discuss
an example that brings together (10a, b). The example, from Solan & Tiersma
2005, is drawn from a real courtroom situation (Bronston v. United States):
6
Christopher Potts
(11)
Comments on Asher and Lascarides
a.
Prosecutor: Do you have any bank accounts in Swiss banks,
Mr. Bronston?
b.
Bronston: No, sir.
c.
Prosecutor: Have you ever?
d.
Bronston: The company had an account there for about six months,
in Zurich.
The literal content of Bronston’s utterances is true: he did not have a Swiss
bank account at the time of this conversation, and the company he mentions
did in fact have one in Zurich for about six months. Bronston was judged to
be dishonest nonetheless. Why? Because he too had once had a Swiss bank
account. Though he never said anything that was literally inconsistent with this
fact, he was perceived to have conveyed its negation. I think this dishonesty
begins, at least, with an IQAP connection: in saying (11d) at this particular
point in this particular discourse, he established an IQAP relationship with the
question expressed in (11c):
(12)
IQAP (11c),(11d)
In the semantics of LoC, (12) has consequences for what Bronston believes about
the relationship between his utterance and the question to which that utterance
attaches. Now, he might try to wiggle out of this by saying that he intended
(11d) to convey a ‘Yes’ answer. This will likely commit him to beliefs that he
does not have — for example, an entailment between the company one works
for having an account at bank X and having an account at X oneself.
Solan & Tiersma (2005) discuss situations like (11) at length, because they
are important for establishing and arguing for subtle instances of perjury. Here
is one of their constructed examples, again from an antagonistic environment:
(13)
John and Mary have recently started going together. Valentino is Mary’s
ex-boyfriend. One evening, John asks Mary, “Have you seen Valentino
this week?” Mary answers, “Valentino’s been sick with mononucleosis
for the past two weeks.” Valentino has in fact been sick with mononucleosis for the past two weeks, but it is also the case that Mary had a
date wih Valentino the night before.
Unless the date was very chaste, this deception is likely to catch up with
Mary. When it does, Julio will likely think back on (13) and decide that Mary
was dishonest. He is unlikely to say it in this way, but I think he would be within
his rights to accuse Mary of falsely leading him to connect two utterances with
IQAP.
7
Christopher Potts
Comments on Asher and Lascarides
What is important about the interactions involving Bronston and the prosecutor, and John and Mary, is that these are noncooperative scenarios. One
speaker is trying to conceal information, and the other is trying to coax it out.
Bronston and the prosecutor are even playing the sort of zero-sum game that
A&L regard as epitomizing conflict. And yet, though this conflict was probably
palpable through the proceedings, speakers still infer IQAP relations.
Thus, there is a tension between the claim that IQAP is rooted in cooperation
and the empirically well-supported idea that it is a sort of automatic, formbased inference. I see two options. We could give up on the derivation from
Cognitive IQAP, or we could give up on the notion that these dialogue games
are not based on cooperativity at the level of communication.
These two approaches make different predictions. If IQAP is completely
form-based, then it should remain powerful even in the most extreme cases of
pure conflict. If it is rooted in cooperativity, then situations in which (2) become
less powerful should also be situations in which speakers are reluctant to draw
IQAP inferences. In such situations, they should insist on direct answers that
leave little or no room for inference.
I think the weight of the evidence points to the idea that IQAP fades in
especially antagonistic circumstances, even when the forms alone would seem
to facilitate such connections. The recent example (14) is typical. Sheehan
is Senator Norm Coleman’s press agent. During the election run-up, Coleman
was accused of accepting gifts from D.C. insiders and not reporting them in the
manner demanded by Senate ethics guidelines. The reporters in this situation
are trying to get Sheehan to respond to allegations that Coleman was given some
fancy suits.4
(14)
REPORTER: On a different subject is there a reason that the Senator
won’t say whether or not someone else bought some suits for him.
SHEEHAN: Rachel, the Senator has reported every gift he has ever
received.
REPORTER: That wasn’t my question, Cullen.
SHEEHAN: The Senator has reported every gift he has ever received.
We are not going to respond to unnamed sources on a blog.
REPORTER: So Senator Coleman’s friend has not bought these suits for
him? Is that correct?
SHEEHAN: The Senator has reported every gift he has ever received.
4 The video is really something to see: http://www.youtube.com/watch?v=VySnpLoaUrI. The
video is titled ‘A Simple Question, “Yes” or “No”??’
8
Christopher Potts
Comments on Asher and Lascarides
REPORTER: Why would say that? Why wouldn’t you give us an answer
yes or no on that?
SHEEHAN: The Senator has recorded every gift he has ever received.
REPORTER: We haven’t asked whether he has recorded every gift he
has ever received and I will take his word that he has recorded every
gift he has received. Has he ever received a gift of suits?
SHEEHAN: The Senator has reported every gift he has ever received.
[Sheehan says ‘The Senator has reported every gift he has ever received’
five more times in two minutes.]
These reporters systematically refuse to regard Sheehan’s “The Senator has
reported every gift he has ever received” as responsive to their questions, though
this is not inherently different from what happens in (4) and (7).
A&L are well aware that the relevant notions of cooperativity are variable in
strength and focus. One sees this in their qualifications to the Grice Cooperative
(GC) element (p. 25). This snippet from a recent email from Alex makes this
even clearer:
cooperativity is a continuum rather than a binary class, and in
many contexts where non-cooperative things are going on people
appear to be behaving cooperatively (and doing so in order to
achieve their ‘non-cooperative’ goals).
I think this is exactly right. I also think it is exactly right to say, as LoC does,
that Cooperativity lies at the heart of IQAP. My conclusion, then, is that
there is a larger role for cooperation than the rhetoric in ‘Commitments’ would
lead one to expect.
4
Beyond IQAP: The nature of enrichment
In the discussion of indirect answers in the previous section, there were really just
two ingredients: (i) the discourse coherence relations (I talked only about IQAP,
but others are relevant), and (ii) inferences from world knowledge, including
knowledge of the context.
Much of the literature on Gricean pragmatics could lead one to believe that
this picture is incomplete — that we still need to sort out the role that the maxims
play in these inferences. I am not sure, though, that there is much more to be
said. I do not know how to make a convincing case for this sweeping statement,
but I can report on some small experiments that might at least convey why I am
willing to stake out this position.
9
Christopher Potts
Comments on Asher and Lascarides
The experiments (simulations, really) are conducted using the software suite
Alchemy (Kok, Sumner, Richardson et al. 2006), which was developed by Pedro
Domingos and his colleagues at the University of Washington.5 Alchemy is a
powerful tool for doing a variety of tasks in machine learning. I think it wasn’t
designed to do pragmatic analysis, but it is suited to it nonetheless.
In general, the set-up works like this: one writes down a series of constraints
in first-order logic. Hard constraints are finished with a period. Soft (defeasible)
constraints are preceded by a real-valued weight and lack a period. Here is an
example adapted from Richardson & Domingos (2006):
(15)
// Friends of friends are friends.
0.7 Friend(x, y) ^ Friend(y, z)) => Friend(x, z)
// Smokers have cancer.
1.5 Smoke(x) => Cancer(x)
// Friends have the same smoking habits.
1.0 Friend(x, y) => (Smoke(x) <=> Smoke(y))
// Rigid facts about the world.
Friend(Ann, Bob)
Friend(Bob, Cal)
Smoke(Ann)
Smoke(Bob)
Ideally, we would learn these weights from data; Alchemy has functionality for
doing just that. Since I am not entirely sure what the training data would need
to be like in the scenarios below, I just stipulate the weights throughout this
commentary.
The situation described in (15) does not contain any direct factual statements
about who has cancer. Nonetheless, it has the right stuff for doing inference
about this predicate. If we ask Alchemy ?Cancer(Ann), it return a relatively
high probability: around 0.8. As we increase the weight on Smoke(x) =>
Cancer(x), this confidence measure increases; in the limit (if this is changed to
a hard constraint), ?Cancer(Ann) is answered positively with a probability of
essentially 1.
Alchemy has efficient methods for searching large spaces of possible worlds
using Monte Carlo methods, so one can make these context files very rich. We
can test intricate situations while at the same time keeping track of all the
assumptions that go into the inferential behavior. The limitation to first-order
logic can be a drag, but the domains are all finite, so it is easy to simulate
second-order expressivity where it is needed (for belief statements, for specifying
what the question under discussion is, and so forth).
In the previous section, I asserted that it is straightforward to ensure that
5 The software and documentation are at http://alchemy.cs.washington.edu/.
10
Christopher Potts
Comments on Asher and Lascarides
Did John pass? is completely answered by He got 60%. Here is an Alchemy
implementation, using only hard constraints, that delivers exactly that result:
(16)
A contextual fragment for (4)
// Generalization.
Score(x,Sixty) => Pass(x).
// Fact about the world.
Score(John, Sixty)
In this context, the chances of John having passed are estimated to be (essentially) 100%. However, if you are not totally sure of Score(x,Sixty) =>
Pass(x), then you might lower the weight on this generalization. This will
inject uncertainty into the extent to which this reply resolves the issue.
With this background in place, let’s turn to a case that naturally creates more
uncertainty about the communicative value of the answer:
(17)
A: Is Sue at work today?
B: She is sick with the flu.
Just what is B saying with this reply? In cooperative situations, A might perceive
B as having given a “No” answer. Then again, suppose that B intends this to be a
complaint about how Sue is going to infect the office. Or suppose that B intends
to mislead, getting Sue into trouble unfairly (as we saw with Bronston and Mary
above).
Different background assumptions deliver different values in these cases. The
generalizations in (18) are intended to capture some general ideas about how
we regard working, being sick, and the connections between those properties.
None could safely be considered a hard constraint, but all have a glimmer of
truth, so we can throw them all in.
(18)
a.
b.
c.
d.
Sick(x) => !AtWork(x)
HardWorking(x) => AtWork(x)
(Malicious ^ Sick(x)) => AtWork(x)
(AtWork(x) ^ Sick(x)) =>
(Malicious(x) v Thoughtless(x))
In Alchemy, !ϕ means ‘not ϕ’. Thus, if (18a) is a hard constraint (or, very heavily
weighted), and we assume that B is truthful, then we achieve a high degree of
confidence that Sue has stayed home:
11
Christopher Potts
(19)
20.0
2.0
1.0
1.0
Comments on Asher and Lascarides
Sick(x) => !AtWork(x).
HardWorking(x) => AtWork(x)
(Malicious ^ Sick(x)) => AtWork(x)
(AtWork(x) ^ Sick(x)) =>
(Malicious(x) v Thoughtless(x))
// Facts.
Sick(Sue)
HardWorking(Sue)
Inference: AtWork(Sue): 0.00005
!Malicious(Sue)
However, if we regard all of these pieces of information as more or less equally
weighted, then Sue’s being hard-working dramatically raises the probability that
she is at work:
(20)
1.0
1.0
1.0
0.5
Sick(x) => !AtWork(x)
HardWorking(x) => AtWork(x)
(Malicious ^ Sick(x)) => AtWork(x)
(AtWork(x) ^ Sick(x)) =>
(Malicious(x) v Thoughtless(x))
// Facts.
Sick(Sue)
HardWorking(Sue)
!Malicious(Sue)
Inference: AtWork(Sue): 0.439906
As we add contextual information, the inferences grow richer. I have not, for
example, included factors pertaining to relative trust levels, possible ulterior
motives, and so forth, nor have I implemented the coherence relations themselves. I see no obstacles to including such constraints, though. My sense is that
implicature behavior, of the sort often attributed to the conversational maxims,
will emerge from such simulations.
5
Other issues we could place under discussion
The central argument of the above commentary is that cooperation is important even in antagonistic discourse relations. My primary evidence is that the
discourse coherence relation IQAP, which is rooted in cooperativity, remains
powerful even when nonlinguistic preferences are misaligned. In addition,
section 4 argues briefly that coherence relations and general inferences from
contextual information can account for a lot of what falls under the heading of
conversational implicature.
Much more could be said, though, about A&L’s provocative, insightful body
of work. I don’t have the time to say much more right now, but I do want to
12
Christopher Potts
Comments on Asher and Lascarides
close by mentioning two issues that are near and dear to me and that A&L’s work
can inform in many ways.
Discourse particles These seem to provide lexical evidence for various discourse relations. They can signal incongruence, surprise, explanation, and a host
of other relationships. Though they are generally optional, discourse conditions
can make them essentially obligatory. There is great potential here for further blurring the distinction between conventionalized meaning and pragmatic
enrichment.
Expressives Expressive language — swears, honorifics, epithets, exclamatives
— takes its toll on the people who hear it, both psychologically and physiologically (Jay 2000; Jay, Caldwell-Harris & King 2008). However, it can also provide
an important release for the speaker, and it can be used strategically, to make
content more memorable, powerful, or relevant (Constant, Davis, Potts et al.
2008). This seems like ripe territory for signaling games of the sort explored by
van Rooy 2003. I can show that expressive signaling systems are stable despite
potentially very high costs to the speaker. It would be fruitful to consider such
systems in the rich, layered logical setting of A&L’s work.
References
Asher, Nicholas & Alex Lascarides. 2003. Logics of Conversation. Cambridge:
Cambridge University Press.
Asher, Nicholas & Alex Lascarides. 2008. Making the right commitments in
dialogue. Paper presented at the University of Michigan Linguistics and
Philosophy Workshop, November 21-23.
Büring, Daniel. 1999. Topic. In Peter Bosch & Rob van der Sandt (eds.) Focus —
Linguistic, Cognitive, and Computational Perspectives, 142–165. Cambridge:
Cambridge University Press.
Constant, Noah, Christopher Davis, Christopher Potts & Florian Schwarz. 2008.
The pragmatics of expressive content: Evidence from large corpora. To
appear in Sprache und Datenverarbeitung.
Groenendijk, Jeroen & Martin Stokhof. 1984. Studies in the Semantics of Questions
and the Pragmatics of Answers. Ph.D. thesis, University of Amsterdam.
Jay, Timothy. 2000. Why We Curse: A Neuro-Psycho-Social Theory of Speech.
Philadelphia/Amsterdam: John Benjamins.
Jay, Timothy, Catherine Caldwell-Harris & Krista King. 2008. Recalling taboo
and nontaboo words. American Journal of Psychology 121(1): 83–103.
13
Christopher Potts
Comments on Asher and Lascarides
Jurafsky, Daniel, Elizabeth Shriberg & Debra Biasca. 1997. Switchboard SWBDDAMSL shallow-discourse-function annotation coders manual, draft 13. Tech.
Rep. 97-02, University of Colorado, Boulder Institute of Cognitive Science,
Boulder, CO.
Kok, Stanley, Marc Sumner, Matthew Richardson, Parag Singla, Hoifung Poon
& Pedro Domingos. 2006. The Alchemy system for statistical relational AI.
Tech. rep., Department of Computer Science and Engineering, University of
Washington, Seattle, WA.
Richardson, Matthew & Pedro Domingos. 2006. Markov logic networks. Machine
Learning 62(1–2): 107–136.
Roberts, Craige. 1996. Information structure: Towards an integrated formal
theory of pragmatics. In Jae Hak Yoon & Andreas Kathol (eds.) OSU Working
Papers in Linguistics, vol. 49: Papers in Semantics, 91–136. Columbus, OH:
The Ohio State University Department of Linguistics. Revised 1998.
van Rooy, Robert. 2003. Being polite is a handicap: Towards a game theoretical
analysis of polite linguistic behavior. In Moshe Tennenholtz (ed.) Proceedings
of TARK 9, 45–58. TARK. URL http://www.tark.org/proceedings/tark_jun20_
03/proceedings.html.
van Rooy, Robert. 2004. Signalling games select Horn strategies. Linguistics and
Philosophy 27(4): 493–527.
Solan, Lawrence M. & Peter M. Tiersma. 2005. Speaking of Crime: The Language
of Criminal Justice. Chicago, IL: University of Chicago Press.
14